System Design Notes

Table of Contents

1. Birthday Paradox & Base62
2. Caching: In-Memory vs Disk
3. CDN with Edge Computing
4. Scaling Strategies
5. Uploading Large Files
6. File Sync Agents
7. File/Data Security
8. Consistency & Transactions
9. Pagination
10. Latency & Caching Strategies

---

Birthday Paradox & Base62

• Collisions are more likely than expected. Only 23 people are needed for a >50% chance two share a birthday.
• For space size N, expect 50% collision chance after about ( \sqrt{N} ) random picks.

Base62 Encoding

Length (power of 62)	Combinations
6	56 billion
7	3.5 trillion
8	218 trillion

---

Caching: In-Memory vs Disk

Speed

Storage	Latency
Memory	100 nanoseconds (0.0001 ms)
SSD	0.1 ms
HDD	10 ms

Throughput

Storage	IOPS (Reads/sec)
Memory	Millions
SSD	~100,000
HDD	100 - 200

---

CDN with Edge Computing

• CDN = geographically distributed PoPs (Points of Presence).
• Edge Computing = logic execution closer to user via Cloudflare Workers or AWS Lambda@Edge.
• Benefits: Reduced latency, faster TTFB, global reach.
• Cautions: Higher cost with scale; limits on memory, execution time.
• Note: Be mindful of cache invalidation and consistency. TTL and CDC can help limit staleness.

---

Scaling Strategies

1. Database:
   • Replication
   • Sharding/Partitioning
   • Backups
2. Services:
   • Horizontal scaling
   • Auto-scaling groups
3. Distributed Cache:
   • Standalone vs Sentinel vs Cluster
4. Counters:
   • Counter batching to reduce network overhead
5. Write Optimization:
   • Batch writes
6. Design Patterns:
   • CQRS (Command Query Responsibility Segregation)
   • Sidecar Pattern
   • Bulkhead Pattern
   • Circuit Breaker

---

Uploading Large Files

• Use Blob/Object storage with pre-signed URLs for direct upload/download.
• Use event triggers (e.g., S3 triggers Lambda) for post-upload actions.
• Prefer Chunking / Multipart upload for:
  • Fault tolerance
  • Parallelism
  • Data integrity (via ETag + Part Number)
• Watch out for:
  • Large file timeouts (e.g., 50 GB on 100 Mbps = ~1.1 hrs)
  • Web server limits (e.g., NGINX default 2GB)
• Consider compression: GZip, Brotli, ZStandard — better for text files.

---

File Sync Agents

Platform	Utility
Linux	fswatch, inotify
macOS	FsEvents
Windows	FileSystemWatcher

---

File/Data Security

1. Encryption at Rest
2. Encryption in Transit
3. Access Control:
   • RBAC (Role-Based Access Control)
   • IAM roles
   • KMS, Vault

---

Consistency & Transactions

Write-centric considerations:

1. Single Database (best)
2. SAGA pattern — with compensating transactions
3. Distributed Lock + 2PC
4. Application-level Lock (e.g., Mutex)

Other notes:

• Use Redis SETNX for distributed locks
• 2PC for cross-shard transactions (may need pre-locking)
• If race conditions rare: use optimistic concurrency (MVCC)
• If possible, normalize schema to a single store to avoid complexity

---

Pagination

Client-Side

• Use lazy loading when dataset is small or partially rendered on demand.

Server-Side

1. Offset Pagination:
   • Simple but problematic under frequent inserts (can cause duplicate/missing rows).
2. Cursor-Based Pagination:
   • Use timestamp + unique ID to avoid duplicates.
   • Monotonic counters also work but prevent page jumps (e.g., page 50).
   • GraphQL Relay uses base64 cursors.

---

Latency & Caching Strategies

Latency mitigation techniques:

1. Caching: Redis, Memcached, CDN
2. TTL: Reduce staleness window
3. CDC: Push updates to cache
4. Geo-Sharding: Serve from nearest region
5. Pooling: Connection/thread pooling
6. Prefetching & lazy loading

Caching Strategies

Strategy	Description
Cache Aside	App reads/writes DB, manages cache manually
Read Through	Cache auto-loads data on miss
Write Through	Writes go to cache and DB simultaneously
Write Back	Write to cache only; DB syncs later (risk of data loss on cache failure)

Eviction Policies:

• LRU (Least Recently Used)
• LFU (Least Frequently Used)
• FIFO (First In First Out)

---

More sections to be added: CAP theorem, Rate limiting, Leader election, Observability, etc.